NATS-Bench
收藏资源简介:
NATS-Bench是由悉尼科技大学计算机科学学院创建的一个统一基准,用于评估几乎所有最新的神经架构搜索(NAS)算法。该数据集包括15,625个拓扑结构候选和32,768个大小结构候选,适用于同时搜索架构的拓扑和大小。NATS-Bench通过在三个不同的数据集上训练每个架构多次,提供了每个架构的训练日志和性能信息,旨在为研究人员提供一个可比较和计算效率高的环境,以开发更好的NAS算法。
NATS-Bench is a unified benchmark developed by the School of Computer Science at the University of Technology Sydney, designed to evaluate nearly all state-of-the-art neural architecture search (NAS) algorithms. This dataset encompasses 15,625 topology candidates and 32,768 size-structure candidates, supporting simultaneous search of both architecture topology and size. By training each architecture multiple times across three distinct datasets, NATS-Bench provides training logs and performance metrics for every architecture. Its core goal is to provide researchers with a comparable and computationally efficient environment for developing improved NAS algorithms.

- 1NATS-Bench: Benchmarking NAS Algorithms for Architecture Topology and Size悉尼科技大学计算机科学学院 · 2021年



